HAL Id: hal-02916854
https://hal.archives-ouvertes.fr/hal-02916854
Submitted on 17 Dec 2020
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Multi-technique physico-chemical characterization of particles generated by a gasoline engine: Towards
measuring tailpipe emissions below 23 nm
Cristian Focsa, Dumitru Duca, J.A. Noble, M. Vojkovic, Y. Carpentier, C.
Pirim, C. Betrancourt, Pascale Desgroux, T. Tritscher, J. Spielvogel, et al.
To cite this version:
Cristian Focsa, Dumitru Duca, J.A. Noble, M. Vojkovic, Y. Carpentier, et al.. Multi-technique physico-chemical characterization of particles generated by a gasoline engine: Towards measur- ing tailpipe emissions below 23 nm. Atmospheric Environment, Elsevier, 2020, 235, pp.117642.
�10.1016/j.atmosenv.2020.117642�. �hal-02916854�
Multi-technique physico-chemical characterization of particles generated by a gasoline engine: towards measuring tailpipe emissions below 23 nm
C. Focsaa,∗, D. Ducaa, J. A. Noblea,1, M. Vojkovica, Y. Carpentiera, C. Pirima, C. Betrancourtb,2, P.
Desgrouxb, T. Tritscherc, J. Spielvogelc, M. Rahmand, A. Boiesd, K. F. Leee, A. N. Bhavee, S. Legendref, O. Lancryf, P. Kreutzigerg, M. Riekerg
aUniversity of Lille, CNRS, UMR 8523 – PhLAM – Laboratoire de Physique des Lasers Atomes et Mol´ecules, F-59000 Lille, France
bUniversity of Lille, CNRS, UMR 8522 – PC2A – Laboratoire de Physico-Chimie des Processus de Combustion et de l’Atmosph`ere, F-59000 Lille, France
cTSI GmbH, 52068 Aachen, Germany
dUniversity of Cambridge, CB2 1PZ Cambridge, United Kingdom
eCMCL Innovations, Castle Park, CB3 0AX Cambridge, United Kingdom
fHORIBA Scientific, 59650 Villeneuve d’Ascq, France
gHORIBA Europe GmbH, Landwehrstrasse 55, D-64293, Darmstadt, Germany
Abstract
Particulate emissions from on-road motor vehicles are the focus of intensive current research due to the impact of the ambient particulate matter (PM) levels on climate and human health. Constant improvement in engine technology has led to significant decrease in the number and mass of emitted PM, but particular concern is raised nowadays by the ultrafine particles. In this context, there is a critical lack of certification procedures for the measurement of the smallest-size (<23 nm) particulate matter emissions. To support the engine development process as well as future certification procedures, a measurement technology for sub-23 nm particles must be designed. The development of a reliable measurement procedure entails understanding the formation and evolution of particles from the engine to the tailpipe via multiple analytical techniques and theoretical simulations.
We present here extensive experimental characterization of ultrafine particles emitted by a gasoline direct injection single-cylinder engine as particle generator. The particles were sampled using a cascade impactor which allows size-separation into 13 different size bins. Chemical characterization of the collected size-selected particles was performed using mass spectrometry, which gives access to detailed molecular information on chemical classes of critical interest such as organosulphates, oxygenated hydrocarbons, ni- trogenated hydrocarbons, metals, or polycyclic aromatic hydrocarbons. Additionally, the morphology of the emitted particles was probed with atomic force (AFM) and scanning electron microscopy (SEM). Tip- Enhanced Raman Spectroscopy (TERS) was applied for the first time to sub-10nm combustion-generated particles to gather information on their nanostructure. The extensive database built from these multiple experimental characterizations has been used as input of a theoretical approach to simulate and validate engine out-emissions. These studies were performed in the framework of the H2020 PEMS4Nano project which aims to the development of a robust, reliable and reproducible measurement technology for particles down to 10 nm for both chassis dyno and real driving emissions (RDE).
Keywords: Internal combustion engine, nanoparticles, physico-chemical characterization, mass spectrometry, emission modeling
∗Corresponding author: cristian.focsa@univ-lille.fr
1Present address: CNRS, Aix Marseille Universit´e, PIIM, UMR 7345, 13397 Marseille cedex, France
2Present address: Ecole Centrale Sup´elec, 91190 Gif-sur-Yvette, France
1. Introduction
1
The particulate matter (PM) is a ubiquitous air pollutant, consisting of a mixture of solid and liquid par-
2
ticles that can remain suspended in the air, ranging from a few minutes up to days or even weeks depending
3
on their size, and hence undergo long-range transport (World Health Organization, 2013). Primary PM can
4
have anthropogenic sources which include PM produced by internal combustion (IC) engines (e.g. on-road
5
vehicles), wear of vehicle components, or industrial activities. Traffic is therefore a significant source of PM
6
emission having a mass size distribution ranging from coarse mode (PM10 <10 µm) to fine mode (PM2.5
7
<2.5µm) particles (Atkinson et al., 2010), and often dominated by submicron particles (e.g. PM0.5<0.5
8
µm). Major improvements in engine technology and the use of alternative fuels over the last few years
9
have helped contain the environmental harm caused by PM emissions(Raza et al., 2018). While NOx and
10
hydrocarbons emissions have been reduced along with the mass and number of emitted particles, one of the
11
downsides has been the shift in the particle diameter towards smaller sizes (lower than 100 nm (Karjalainen
12
et al., 2014)), thus likely contributing to air pollution,i.e. a public health issue (e.g. Manke et al. (2013);
13
Sager and Castranova (2009); Seaton et al. (2009); EPA (2009)). The G-20 countries account for 90% of
14
global vehicle sales, and 17 out of the 20 members have chosen to follow the European regulatory pathway
15
for vehicle emissions control (Williams and Minjares, 2016). Therefore, information about the particle num-
16
ber (PN) of ultrafine particles is becoming more and more valuable for vehicle certification. Specifically,
17
sub-23 nm particles have recently attracted a lot of attention for mainly two reasons. For one, sub-23 nm
18
particles can be produced, and sometimes in high concentrations, in both diesel and gasoline direct injection
19
(GDI) engines (e.g. Giechaskiel et al. (2014)). Second, the harmfulness of the particles has been shown
20
to correlate better with surface area than with mass (Donaldson et al., 1998; Oberdorster, 1996), which
21
becomes important for ultrafine particles even though their residence time in the atmosphere is shorter.
22
However, it has been estimated that the percentage of sub-23 nm solid particles that is not measured by
23
current certification procedures (that have a cut-off size of 23 nm) could reach 30-40% of the total PN for
24
gasoline vehicles utilizing direct injection, and be potentially higher when alternative fuels are being used
25
(Giechaskiel et al., 2017). Therefore, the critical lack of certification procedures for the measurement of
26
ultrafine (<23 nm) particulate matter emissions should be addressed. Current efforts (Kontses et al., 2020;
27
Chasapidis et al., 2019; Lee et al., 2019) focus on providing solid scientific ground to allow lowering the
28
23 nm limit to 10 nm, with the aims of providing robust particle number measurement methodology and as-
29
sociated instrumentation. Although possible in principle (at least in well-controlled laboratory conditions),
30
going below this limit appears as very challenging as nanometer-size particles raise important sampling,
31
measurement and quantification issues which can result in undesired biases and artifacts (Simonen et al.,
32
2019). This difficulty is of course enhanced when designing reliable portable measurement systems dedicated
33
to monitoring nanoparticle emissions in real-driving conditions.
34
The development of a robust and reliable certification procedure for measuring even small particles
35
requires a good understanding of the characteristics of the emitted particles. There are several studies that
36
examine particle emissions of modern direct injection engines during laboratory tests (chassis dynomometer)
37
and even under real driving conditions. The number, mass, and size distribution of emitted particles have
38
been already studied for a wide variety of engine operating conditions (Khalek et al., 2010; Baral et al.,
39
2011; Maricq et al., 2011; Karjalainen et al., 2014; Vojt´ıˇsek et al., 2014; Momenimovahed et al., 2015; Ko
40
et al., 2019). It was also shown that engine emissions tend to be optimized for the certification test cycle,
41
while other regimes are often overlooked (Vojtisek-Lom et al., 2009; Weiss et al., 2011). As a result, engine
42
emissions in real driving conditions (RDE) tend to be higher (Kayes et al., 2000; Kristensson et al., 2004).
43
The chemical composition of exhaust particles was also examined and enabled the estimation of fuel and
44
oil contribution to elemental (EC) and organic carbon (OC) (Kleeman et al., 2008; Fushimi et al., 2016;
45
An et al., 2016). However, there is still very little information about the chemical composition of fine and
46
ultrafine particles emitted by modern engines and their origin. The latter is especially important for the
47
development of a Particle Measurement Program (PMP) capable of measuring ultrafine particles, below
48
23 nm (Zheng et al., 2011).
49
To explore the physico-chemical characteristics of the emitted particles, it is necessary to perform a multi-
50
technique analysis of GDI engine emissions. The objective of this study is to build a large database through a
51
thorough size-dependent physico-chemical characterization of a wide variety of combustion-generated parti-
52
cles sampled under different engine operating conditions. The database includes extensive information about
53
the size-dependent molecular-level chemical composition of emitted particles as small as 10 nm obtained with
54
high-performance mass spectrometry techniques. Recently developed advanced statistical methodologies (Ir-
55
imiea et al., 2018, 2019; Duca et al., 2019), based on principal component analysis (PCA) and hierarchical
56
clustering analysis (HCA), are used in this work to highlight subtle differences, as well as similarities, between
57
different-sized particles. These data are complemented by the particle morphology information obtained with
58
a scanning electron microscope (SEM) and an atomic force microscope (AFM), while unprecedented tip-
59
enhanced Raman spectroscopy (TERS) measurements on individual combustion-generated particles down
60
to a few nm are performed to gather insights about their nanostructure. Engine set-points used for off-line
61
characterization of particles were selected using laser induced incandescence (LII) – an on-line measure-
62
ment technique able to provide real-time information on the PM size-distribution and volume fraction. The
63
obtained database is then utilized in a theoretical approach (referred to as Model Guided Application –
64
MGA) which combines detailed physico-chemical simulation together with advanced statistical techniques
65
of parameter estimation, computational surrogate generation and sensitivity analysis (Lee et al., 2019). The
66
MGA aims to predict the particle formation in the engine, as well as particle dynamics in the exhaust system
67
and sampling line. Although the modeling work presented in this paper is related to a single-cylinder test
68
engine (used as a particle generator), the knowledge gained here (experimental and theoretical) can be easily
69
transferred to multi-cylinder engines and even expanded to drive cycle simulations, thus providing crucial
70
information for the development and optimization of particle emission monitoring systems (PEMS). These
71
studies were carried out in the framework of the H2020 PEMS4Nano project (www.pems4nano.eu) where a
72
bottom-up approach was adopted in order to develop a robust, reliable and reproducible particle emission
73
monitoring system (PEMS) for both chassis dyno and real-driving conditions.
74
2. Materials and Methods
75
2.1. Single cylinder engine and sampling line
76
The particulate matter investigated in this work was produced by a generic gasoline direct injection
77
single-cylinder test engine, whose specifications are given in Table 1. The engine was operating on Euro
78
Stage V E5 Gasoline (CEC-RF-02-08 E5). The temperature for both coolant and oil (Agip SIGMA, 10W-
79
40) was set to 80°C. All the operating points studied are described in Table 2 (e.g. applied load, injection
80
timing). For all the set-points, the engine was operating at 2000RPM with the ignition timing set for the
81
maximum brake torque (MBT). Between each operating point, the engine was operated with methane (CH4)
82
for 30 minutes to ensure the cleanliness of the combustion chamber.
83
Table 1: Engine specifications (b/aTDC – before/after Top Dead Center)
Specification Value
Cylinder head Pentroof type
Compression ratio 12.5:1
Bore 82 mm
Stroke 85 mm
Stroke volume 449 cm3
Fuel direct injection system Central mounted generic six-hole injector Injection pressure 150 bars
Spark plug location Exhaust side Intake valve timing: Open 334 deg. bTDC
Close 166 deg. bTDC Exhaust valve timing: Open 154 deg. aTDC Close 330 deg. aTDC
Monitoring and sampling of particles from the generic single-cylinder test engine was performed using
84
a home-built sampling line. A partial flow was taken from the exhaust duct approximately 5 cm after
85
the exhaust port and was then supplied via 8 cm long transfer line to DEKATI FPS 4000 (Fine Particle
86
Sampler) two-stage dilution system (1:3 and 1:10 dilution ratios for the hot (180°C) and cold (35°C) dilution
87
stages respectively). The rest of the exhaust flows through an optical cell for on-line soot monitoring by LII.
88
As detailed below, the LII technique was used for preliminary characterization and selection of the engine
89
operating conditions. Additionally, the in-cylinder pressure profile and gas phase emissions (CO,µHC and
90
NOx) were monitored.
91
Figure 1: Schematic representation of the sampling line. Sampling with NanoMoudi on Al foils can be done with or without the catalytic stripper (dashed blue rectangle).
After dilution, the flow of particles was split into three branches. On the first one, an Engine Exhaust
92
Particle Sizer (EEPS; TSI, model 3090) was used to measure particle number and size distribution. A
93
combination of a Nano-Differential Mobility Sizer (Nano-DMA; TSI, model 3085) and a Nanometer Aerosol
94
Sampler (NAS; TSI, model 3089, Fig. 1), placed on the second branch, was used to collect particles for
95
structural and morphological analyses, thus enabling sampling within very narrow size bins. Finally, the
96
sampling for chemical analysis was conducted on the third branch with a NanoMOUDI II cascade impactor
97
(TSI/MSP, model 125R). Being a “Micro-Orifice Uniform-Deposit Impactor”, the NanoMoudi works on the
98
well-known principle of inertial impaction (Marple et al., 1991; Marple, 2004). The MOUDI impactors have
99
sharp stage characteristics with steep cut-offs and low inter-stage wall losses. The used model comprises 13
100
stages, with nominal cut-off sizes of 10000, 5600, 3200, 1800, 1000, 560, 320, 180, 100, 56, 32, 18, and 10
101
nm. Particles were deposited on uncoated aluminum foils (heated for 12 hours in an oven at 300°C prior to
102
the sampling).
103
A catalytic stripper (CS) can be placed before the NanoMoudi in order to investigate its impact on
104
the properties of the particles sampled in several engine regimes. The CS was built by the University of
105
Cambridge in a similar fashion to the commercially-available models (Catalytic Instruments). The coating
106
was commercially available precious metal loading (100–300 g/m2) with an alumina wash coat. The CS was
107
calibrated and monitored with a temperature controller to ensure that the centerline temperature of the
108
aerosol achieves 350°C prior to entering the catalytic monolith contained within the device. The catalytic
109
stripper performance in terms of hydrocarbon removal was tested to be >99% for 30 nm tetracontane
110
particles at a concentration of >104 cm−3. The precise catalyst chemistry has not been measured. The
111
penetration efficiency was >60% for 10 nm particles and the entire system residence time was 1–2 s. The
112
influence of the catalytic stripper on particle characteristics was studied since its usage in a PMP compliant
113
system and is preferred over an evaporative tube (ET) as it significantly reduces the number of artifact
114
ultrafine particles formed by renucleation of semivolatiles, thus considerably increasing the reliability of the
115
system (Zheng et al., 2011). The stripper was operating at 370°C and was preheated prior to the sampling.
116
The off-line characterization techniques used for structural and morphological analysis (described in
117
section 2.3.1) require samples with a low coverage,i.e. isolated individual particles. At the same time, the
118
chemical characterization (Section 3.3) requires more loaded samples, ideally with a homogeneous layer of
119
particles fully covering the deposition substrate. Consequently, the sampling time for the samples intended
120
for chemical characterization was much longer. Moreover, to obtain similar coverages, the time was optimized
121
for each engine operation regime, the longest (12h) being for the NHC regime and the shortest (6h) for
122
theEMone, respectively (Table 2). Regarding the samples intended for the structural and morphological
123
analysis, the collection time varies between 30 minutes and 1h, depending on engine conditions.
124
Table 2: Studied engine operating points (IMEP – Indicated Mean Effective Pressure,λ– ratio of the actual air/fuel ratio to stoichiometric). “Optimal engine conditions” refer to optimal injection and ignition timings set for MBT.
Set-point IMEP, bar λ Injection
Comments
±0.01 deg. BTDC
NS 3 1.01 270
NM 6 1.01 270
NH 12 1.01 270
NHC 12 1.01 270 With a catalytic stripper
OM 6 1.01 270 Addition of oil
EM 6 0.70 270 Low air/fuel ratio
EI1 12 1.01 305 Early injection
EI2 12 1.01 311 Early injection
“N” in the set-point designations stands for normal engine operation (optimal engine condi- tions); “S/M/H” – small/medium/high load applied to the engine; “C” – catalytic stripper;
“OM”, “EM” – malfunction regimes (“O” – addition oil; “E” – excess of fuel resulting in low air/fuel ratio);
“EI” – optimal air/fuel ration, high load, early injection.
2.2. On-line characterization by Laser-Induced Incandescence
125
LII is a laser-based technique allowing soot concentration monitoring in various environments, for instance
126
in flames (Schulz et al., 2006) and in the combustion chamber of internal combustion engines (Boiarciuc
127
et al., 2007; de Francqueville et al., 2010; Koegl et al., 2018). LII also proved to be suitable to control
128
real time soot emission directly in the exhausts of diesel (Case and Hofeldt, 1996; Bachalo et al., 2002),
129
gasoline (Smallwood et al., 2001; Bardi et al., 2019), and aeronautical engines (Delhay et al., 2009). In
130
this work, the objective of the LII measurements is to study the evolution of the soot volume fraction with
131
engine conditions in order to select the appropriate engine set-points for off-line soot-characterization. The
132
principle of LII consists first in heating combustion-generated particles up to 3500-4200K, generally using a
133
pulsed laser, and then recording the radiation associated with the thermal emission while they cool down.
134
This incandescence emission follows the Planck law and depends on the emission properties of particles at
135
the detection wavelength. LII is mainly used to determine the soot volume fraction (fν) and the primary
136
particle diameter (Schulz et al., 2006). Soot volume fraction can be determined from the time-resolved LII
137
peak signal as follows:
138
fν ∝
LII
E(mλ)·B(T, λ)
(1) whereB(T, λ) is the Planck function,T is the peak reached temperature,λ- the emission wavelength,mλ
139
- the refractive index of the particle andE(mλ) - soot absorption function.
140
The decay of the LII signal is related to the cooling time of particles and therefore to their size. Primary
141
particle diameter can hence be determined, provided a model of the energy transfer during laser heating
142
is available (Michelsen et al., 2007; Betrancourt et al., 2017). The LII technique has been continuously
143
developed and optimized at University of Lille during the last decades (Bladh et al., 2006; Bejaoui et al.,
144
2014). It has been applied to, e.g., the measurement of the soot volume fraction in turbulent flames
145
(Lemaire et al., 2010) or in the exhaust of aero-engines (Delhay et al., 2009). Recent efforts were devoted
146
to the detection of very small soot nanoparticles (Betrancourt et al., 2017, 2019). The LII technique was
147
adapted here to probe soot particles in an optical cell mounted on the exhaust duct (Fig. 1), simultaneously
148
with PM collection for off-line analyses. The choice of the optimum cell location took into consideration
149
the layout and space constraints of the test bed. Details of the LII experimental set-up are provided in the
150
Supplementary Fig. S1. Briefly,LII measurements were performed with an excitation wavelength of 1064
151
nm generated by a Nd:YAG pulsed laser (Quantel Brilliant, 10 Hz repetition rate, 5 ns pulse duration). An
152
optical attenuator is used to adjust the energy of the laser between 0 and 350 mJ/pulse. The excitation
153
laser beam has a Gaussian profile and a diameter of 6 mm. A laser fluence of 0.32 J/cm2 was used for
154
excitation, thus maximizing the LII signal and limiting its dependence on laser fluctuations. The LII signal
155
is collected by two achromatic lenses (f1=200 mm, f2=100 mm) and focused on the collection slit (100 mm x
156
0.2 mm). The signal, in the 400-825 nm spectral range, is then detected with a photomultiplier tube (PMT)
157
(Hamamatsu R2257) and recorded by an oscilloscope (LeCroy 6050A, 500 MHz bandwidth, 5 GS/s sampling
158
rate) triggered with a photodiode (Hamamatsu S1722-02). The PMT is equipped with a short-pass filter
159
with a cut off at 825 nm. The performance of the set-up (with exactly the same geometry and collection
160
volume) was priorly optimized and evaluated in the laboratory of Lille using carbon particles produced
161
by a PALAS DNP 2000 generator. Set-points providing nearly constant particles concentrations could be
162
obtained in the range 105 to 1.2·106 particles/cm3. From these measurements a limit of detection of 104
163
particles/cm3 was established.
164
2.3. Off-line characterization
165
2.3.1. Morphology and structure
166
For structural and morphological studies, particles collected on Au-coated or bare Si wafers are analyzed
167
with Scanning Electron Microscopy (SEM) and Atomic Force Microscopy / Tip Enhanced Raman Spec-
168
trometry (AFM/TERS). The SEM Merlin (Zeiss) is used for the morphology analysis of collected particles;
169
it features a spatial resolution of 1 nm and high detection efficiency. For the study of ultra-fine particles, a
170
low accelerating voltage (1kV) was chosen to decrease the size of the interaction volume and improve the
171
spatial resolution. Moreover, it enables the study of ultra-fine combustion-generated particles without the
172
need of a conductive coating. Additionally, SEM images are used to determine the surface coverage and
173
adjust, if needed, the sampling time.
174
The AFM (SMART SPM, HORIBA Scientific) features high-resolution cantilever-based tips (Hi’Res-
175
C14/Cr-Au,µmasch, typical 1 nm-radius) for a high lateral resolution. A NanoRaman system that combines
176
a SMART-SPM (HORIBA Scientific) with a Raman spectrometer (LabRAM HR Evolution Nano, HORIBA
177
Scientific) is used for tip-enhanced Raman measurements (TERS). The system is based on a reflection
178
configuration allowing the use of a 100x objective lens (NA 0.7) with a 60° angle. The incident laser
179
(633 nm, p-polarized) is focused through the objective onto the apex of the cantilever-based silver TERS
180
probe (Ag coated OMNI TERS probe). The collection of the back-scattered signal is performed through the
181
same objective. The reliable stability of the whole system allows to maintain the laser-tip alignment and
182
accurate XYZ position during the TERS map acquisition (at least 1 hour).
183
2.3.2. Chemical composition
184
Two-step laser mass spectrometry
185
Our group has extensively developed the two-step laser mass spectrometry (L2MS) technique over the last
186
decade to specifically probe the chemical composition of combustion byproducts (Popovicheva et al., 2017;
187
Delhaye et al., 2017; Faccinetto et al., 2011, 2015; Moldanov´a et al., 2009). The main advantages of L2MS
188
are its high sensitivity and selectivity with regards to specific classes of compounds owing to the resonant
189
ionization processes that can be adapted, by using different ionization wavelengths, thus targeting different
190
classes of compounds and reaching sub-fmol limit of detection,e.g. for PAHs (Faccinetto et al., 2008, 2015).
191
Systematic studies (Mihesan et al., 2006, 2008) have led to the optimization of the laser desorption process
192
(and its coupling with the subsequent ionization step (Faccinetto et al., 2008), which ensures a soft removal
193
(with minimum internal excess energy) of molecules adsorbed on the particle surface, and thus avoids/limits
194
both their fragmentation and the in-depth damaging of the underlying carbon matrix (Faccinetto et al.,
195
2015). In this work, a new mass spectrometer (Fasmatech S&T) is used, combining ion cooling, Radio
196
Frequency (RF) guiding and Time of Flight (ToF) analyzer, which allows us to reach a mass resolution of
197
m/∆m ∼15 000. In this new experimental setup, the sample, placed under vacuum (10−8 mbar residual
198
pressure), is irradiated at 30° angle of incidence by a frequency doubled Nd:YAG laser beam (Quantel
199
Brilliant,λd= 532 nm, 4 ns pulse duration, 100 mJ cm−2 fluence, 10 Hz repetition rate) focused to a 0.07
200
mm2 spot on the surface. The desorbed compounds form a gas plume expanding in the vacuum normally
201
to the sample surface, and are ionized by an orthogonal UV laser beam (Quantel Brilliant, λi= 266 nm,
202
4 ns pulse duration, 10 Hz repetition rate, 0.3 J cm−2 fluence). At this ionization wavelength, a high
203
sensitivity is achieved for PAHs through a resonance enhanced multiphoton ionization process 1+1 REMPI
204
(Zimmermann et al., 2001; Haefliger and Zenobi, 1998; Thomson et al., 2007). The generated ions are then
205
RF-guided to a He collision cell for thermalization and subsequently mass analyzed in a reflectron time of
206
flight mass spectrometer (ToF-MS). A high sensitivity for aliphatic compounds was achieved with a similar
207
instrument where the desorbed plume is ionized by a single photon ionization (SPI) process atλi=118 nm.
208
The ninth harmonic of the Nd:YAG laser is generated in a coherent nanosecond source by frequency-tripling
209
of a 355 nm pump beam (Continuum Surelite, 10 ns, 10 Hz) in a low-pressure Xe cell (Popovicheva et al.,
210
2017).
211
Secondary Ion Mass Spectrometry
212
The particles were also analyzed with a commercial IONTOF TOF.SIMS5secondary ion mass spectrometer
213
(SIMS) with maximum resolving power ofm/∆m ∼10 000. The sample is placed on a holder and introduced
214
into a vacuum chamber with residual pressure of∼10−7mbar. The surface of the sample is probed by a Bi+3
215
pulsed ion beam at 25 keV. The primary ion source delivers a pulsed current of 0.3 pA. The penetration
216
depth of the primary ions is typically 1-3 nm. In the static mode, used for the surface molecular analysis,
217
the ion dose is limited to a level at which every primary ion should always hit a fresh area of the sample. A
218
small fraction of the ejected atoms/molecules are ionized (secondary ions) and can thus be analyzed using a
219
time of flight tube (V mode). Mass spectra are recorded in both positive and negative polarities, to obtain
220
the maximum amount of information on the sample (Irimiea et al., 2018, 2019).
221
2.4. Theoretical model
222
The model-based workflow (Lee et al., 2019) simulates the in-cylinder gas phase and particulate produced
223
in the engine and up to the sampling stage. There are two stages in the workflow: i) calibration of the in-
224
cylinder pressure and engine-out emissions using a Stochastic Reactor Model (SRM), and ii) investigation
225
on the formation of organic carbons in the sampling stage using a reactor network model.
226
SRM Engine Suite for in-cylinder combustion
227
The SRM is described by the probability density function (PDF) transport equation and it is assumed
228
that the inhomogeneity is described by the distribution represented by the PDF everywhere. The multi-
229
dimensional PDF transport equation is represented by an ensemble ofNparstochastic parcels. Each stochas-
230
tic parcel is described by a collection of gas phase species mass fraction and a temperature, i.e.
231
(Y1(i), . . . , YN(i)
S, T(i)) (2)
where the superscript indices are labels for the parcels (up to Npar) and NS is the number of species
232
considered in the gas phase chemical mechanism. These quantities evolve through a number of processes,
233
e.g. chemical reaction, turbulent mixing, flame propagation, wall impingement etc. (Bhave and Kraft, 2004;
234
Etheridge et al., 2011; Smallbone et al., 2011; Lai et al., 2018). More detailed information on the application
235
of the SRM in a GDI context can be found in Lee et al. (2019).
236
In addition to the gas phase composition and temperature, each parcel contains a particle population
237
that consists of soot and inorganics. This model is directly coupled with the gas phase chemistry and a
238
three-dimensional type-space is used to describe the particles:
239
P = (m, nprim, YSOF) (3)
where mis the particle’s mass, nprim is the number of primary particles andYSOF is the mass fraction of
240
soluble organic fractions (SOF). The particle population evolves through inception from aromatics from the
241
gas phase, surface growth, oxidation, and aggregation. The sectional method is used to solve the population
242
balance model.
243
Reactor network model for the dilution stage
244
The engine-out composition, both gas and particulate phases, is passed onto a reactor network model to
245
mimic the dilution process in the Dekati FPS 4000 diluter described in Section 2.1. The reactor network
246
model is illustrated in Fig. 2. R1 and R3 represent the hot (180°C) and cold dilution (35°C) stages
247
respectively, whereas R2 and R4 represent the piping connections. The model simulates a population of
248
particles with the type space described by Eq. 3. From here, they can be post-processed to give particle
249
size distributions, PN, PM and size-resolved SOF mass fractions. The outputs from the reactor network
250
model are then compared with measurement data, which include particle size distributions and size-resolved
251
chemical particle characterization.
252
Figure 2: Reactor network model for the Dekati FPS 4000 diluter.
3. Results and Discussion
253
3.1. LII measurements within the exhaust duct
254
The LII technique was used in this work to monitor the evolution of the soot volume fraction (proportional
255
with the LII peak signal, see section 2.2) upon various engine regimes. Fig. 3a shows the evolution of the
256
LII signal (fν) with five engine set points. One can clearly see that the main parameter inducing a strong
257
effect on the soot volume fraction is the timing of injection, leading to a 150-fold increase between theNH
258
(270 deg. BTDC, see Table 2) and theEI2 (311 deg. BTDC) regimes. This trend is correlated with the
259
occurrence of pool fires (i.e. liquid fuel burning on the top of the piston) observed only for the points
260
EI1 and EI2 using endoscopic imaging of the combustion chamber. This observation is consistent with
261
the ones revealed by Velji et al. (2009) who experimentally identified that the main source of soot in the
262
homogeneous combustion in a gasoline engine is due to the pool fires, while in the stratified mode the soot
263
may also originate from local rich regions in the combustion chamber. Bardi et al. (2019) also observed
264
a correlation between the occurrence of pool fires and the large increase in soot particle number. On the
265
other hand, a moderate increase (by a factor of 3.2) is recorded when the indicated mean effective pressure
266
(IMEP) quadruples, from 3 (NS) to 12 bar (NH). This is consistent with the work of Wang et al. (2014)
267
who suggested that this trend could be due to slightly longer combustion duration for a clean injector.
268
In addition to the control of the soot volume fraction, time-resolved LII measurements can provide
269
qualitative information about the size of the emitted particles. The LII signal time decays exhibit significant
270
variations depending on the engine regime (see Fig. 3b). The “normal regimes” (NS, NM, NH) show
271
similar decay times (i.e. relatively low influence of the IMEP on the particle size), whereas the non-optimal
272
engine operation conditions (early injection, EI1 and EI2) clearly show longer decay times i.e. larger
273
primary particle sizes. The interpretation of these time-resolved LII decays in terms of soot primary particle
274
diameter requires complex energy transfer modeling (Michelsen et al., 2007) which is beyond the scope of
275
this work. This modeling is made even more difficult because size distributions of the particles are expected
276
to be bimodal. Wang et al. (2014) measured on-line the size distributions of the particles in different GDI
277
conditions by scanning mobility particle sizing (SMPS) and highlighted that most of the electrical mobility
278
diameter distributions were bimodal. Particularly they showed that the smallest nucleation mode particles
279
could be prevalent in some engine conditions. Providing that LII modeling is undertaken, the coupling of
280
LII with EEPS or SMPS offers promising perspectives. Indeed, in case of bimodality (which can be revealed
281
by SMPS or EEPS), the LII could provide the diameter of the largest primary particles. The extraction of
282
the primary particle diameter from the LII decays, even biased towards the largest ones, could give real-time
283
access to this important characteristic without resorting to the use of off-line analyses (by e.g., TEM).
284
Figure 3: (a) LII peak signal (a.u.) measured at different engine set-points; (b) time-resolved LII signal.
These observations provide precious information for optimizing the simultaneous sampling of the par-
285
ticulates for off-line analyses. Further efforts will be devoted to calibratefν using an extinction technique
286
at a given engine set-point (Betrancourt et al., 2019) and to combine LII with particle sizers as mentioned
287
above in a context of very small particles (down to 10 nm). From the LII measurements, several set-points
288
showing a high stability have been selected for the off-line characterization, Table 2.
289
3.2. Structure and morphology
290
A significant challenge in collecting and studying combustion-generated particles is avoiding the agglom-
291
eration of the smallest ones during sampling. As small particles are the focus of this work, proper sampling
292
is crucial to avoid the aggregation of small particles that would result in their collection on a different im-
293
paction stage (with a higher cut-off size compared to the dimensions prior to aggregation). Several samples
294
of particles are analyzed with SEM, Fig. 4, in order to ensure that our experimental setup is able to sample
295
and collect small primary particles. The particles are collected on Au-coated Si wafers with the Nano-DMA,
296
set to a mobility diameter of 18±5 nm (see Section 2.1).
297
One can see that the samples (obtained in different engine regimes) contain predominantly single, spher-
298
ical particles with a diameter ranging from 17 to 23 nm, Fig. 4, and are presumably primary soot particles.
299
This is in agreement with previous studies on the emitted particles from GDI engines (Gaddam and Wal,
300
2013; Price et al., 2007). The fact that primary particles were successfully sampled indicates that the ex-
301
perimental setup is working as intended, minimizing the aggregation of particles to an extent that should
302
not affect the subsequent chemical analysis.
303
AFM analysis is performed on samples collected in the NH engine regime (optimal conditions, high
304
load), Table 2. The technique is used to determine the distribution and the size of single particles. AFM
305
measurements are performed on a 1.2 x 1.2µm region (300 x 300 pixels), Fig. 5a. The heights of particles 1
306
and 2 were determined to be 10 nm and 6 nm, Fig. 5d. The measured diameter is highly dependent on the
307
used tip, especially when the radius of the scanned object is close to the one of the tip. The Full Width at
308
Half Maximum (FWHM) diameters of particles 1 and 2, obtained with a sharp tip (r=1 nm), are <48 nm
309
and<43 nm, respectively.
310
Two smaller zones have been characterized by TERS with a resolution of 7 nm/pixel (zone I) and 5
311
nm/pixel (zone II) and with a Raman integration time of 100 ms (one spectrum/pixel). The TERS maps
312
of zones I and II are presented in Fig. 5b and 5c, respectively. Raman spectra of carbon materials contain
313
two main features (see Fig. S2 in Supplementary Material): the G band, derived from in-plane motion of
314
carbon atoms (around 1580 cm−1) and the D band (D1), attributed to lattice motion away from the center
315
Figure 4: SEM images of size-selected particles (size bin centered at 18 nm) collected with a NAS in theNH(left panel) and NM(right panel) engine operating regimes.
of the Brillouin zone (around 1270 – 1450 cm−1). The TERS maps displayed in Fig. 5b,c are obtained by
316
integrating the G band only.
317
Figure 5: AFM and TERS study of nanoparticles collected in theNHengine regime (optimal conditions, high load). Topog- raphy image (1.2 x 1.2µm) (a), TERS mapping of zone I, particles 1 and 2 (b), and zone II, particles 3 and 4 (c), topographic cross-section of particles 1 and 2 (d).
Besides TERS maps, TERS spectra were obtained for particles 2, 3, and 4 and are displayed in Fig. S2.
318
The three spectra (Fig. S2) exhibit very different profiles (for example, the relative intensity of the D band
319
to the G band is higher for particle 2 and lower for particles 3 and 4) which is indicative of major structural
320
variations between particles collected in the same engine regime and the same size bin. A more quantitative
321
approach involves a five-band fitting method, often applied to Raman spectra of soot particles (Sadezky
322
et al., 2005; Parent et al., 2016), in which the G, D1, D2, and D4 bands are fitted with a Lorentzian profile,
323
whereas the D4 band is fitted with a Gaussian profile. The fit results are presented in Fig. S2. The spectral
324
parameters determined by curve fitting are highly variable across the three particles (Table S1), confirming
325
our first observation. For instance, the ratio of the integrated areas of D1 to those of (G+D2) bands is
326
usually considered as a good indicator of the order in the soot structure (Carpentier et al., 2012). This value
327
increases from 1.5 (particle 4) to 2.5 (particle 3) and to 3 (particle 2), suggesting that particle 2 has a much
328
more disordered structure compared to particles 3 and 4. In addition, the presence of strong signatures at
329
1208 and 1278.6 cm−1 on the spectra of particles 3 and 4 could be an indicator of organic molecules present
330
on the surface of the particles.
331
These TERS measurements, performed (to the best of our knowledge) for the first time on ultra-fine
332
combustion-generated particulate matter, reveal the diversity (at least in terms of nanostructure) of soot
333
particles generated by the same engine, operating in the same conditions, and collected in the same size
334
bin. This observation is of paramount importance for the interpretation of our off-line chemical composition
335
measurements discussed in section 3.3, which, conversely, do not probe individual particles but rather a
336
high number of them because of the diameter of the laser or ion beam used in the experiments. Therefore,
337
the properties and trends we derive in section 3.3 are averaged over many (different) particles making them
338
statistically significant. These trends offer valuable input for the theoretical model, also operating with
339
statistical approaches on particles ensembles.
340
3.3. Chemical characterization
341
To obtain a comprehensive chemical characterization and thus provide valuable inputs for building and
342
validating the theoretical model, size-selected particles are analyzed with L2MS and SIMS. Data-treatment
343
was performed according to our recently developed comprehensive methodology (Irimiea et al., 2018, 2019;
344
Duca et al., 2019) which includes powerful statistical techniques such as PCA and HCA. Since the particles
345
emitted in a certain engine regime are size-selected, a notation scheme that indicates the size of collected
346
particles, in addition to the engine regime will be used from here forth (<engine regime>:<upper size
347
limit>-<lower size limit>),e.g. OM:180-100. Due to a low particle coverage, the substrate (Al foil) was
348
also probed which led to the formation of Al+ and Al2O+ ions, detected at m/z 27 and 70, respectively.
349
Consequently, even though the presence of Al compounds in such samples is typically a marker of engine
350
wear, these peaks will rather be associated with the aluminum substrate in our case.
351
3.3.1. Size discrimination
352
One of the major points that need to be determined is whether the chemical composition of emitted
353
particles varies with their size. On that ground, size-selected particles produced in two different engine
354
regimes (OM– optimal conditions with an addition of oil, andEM– low air/fuel ratio) have been analyzed.
355
These particular engine regimes were chosen as they simulate two completely different malfunctions, i.e.
356
extreme cases of operation, equivalent to either a “worn out” engine with high oil consumption (OM) or
357
a fault in the fuel system whith a normal oil consumption (EM). This information, along with similar
358
analyses performed on size-selected particles emitted in other operation conditions, allows for a more precise
359
calibration of the MGA.
360
“Worn out” engine (OMregime)
361
The analysis of samples obtained in the OM regime (optimal conditions, addition of oil) was performed
362
with two-step laser mass spectrometry with two different ionization wavelength (266 nm and 118 nm), thus
363
enabling us to selectively target either aromatic or aliphatic compounds. Several metals were detected (Na+,
364
K+ and Ca+) with both ionization wavelengths. While Na+ and K+ are mostly associated with fuel, they
365
can also be present in the lubricating oil as trace elements (Huang et al., 1994; Cadle et al., 1997; Cross
366
et al., 2012; Dallmann et al., 2014). In addition, the origin of Ca+ is most likely the lubricating oil, since
367
Ca is a component of detergent additives, widely used in modern motor oils (Cadle et al., 1997; Dallmann
368
et al., 2014). Mostly aromatic species are detected when the ionization of laser-desorbed compounds is
369
performed at 266 nm (Fig. 6). They are visually separated into two groups: one of lower and one of higher
370
mass compounds. The first group is comprised of aromatic species with one to two aromatic rings and their
371
alkylated derivatives (m/z 78 – 170) and is present in the spectra of all the samples with varying relative
372
intensities. The second group is composed of aromatic compounds with three and more rings. According
373
to Bari et al. (2010), the first group can be considered as volatile species, as they are mostly present in the
374
gas phase. The second group is comprised of semi and non-volatile compounds. Therefore, the intensity
375
ratio of the first to the second group is related to the overall volatility of the organic layer on the surface
376
of particles. Species with masses betweenm/z 178 and 398 are detected on larger OMparticles (180 – 32
377
nm),i.e. on samplesOM:180-100,OM:100-56, andOM:56-32, while for the sampleOM:32-18 (32 –
378
18 nm) the highest detected mass is only m/z 278. Table S2 displays the assignment of the most intense
379
mass peaks recorded by L2MS with two different ionization wavelengths (256 nm and 118 nm).
380
The intensity ratio of species belonging to the second group to those in the first one changes from sample
381
to sample, within the OM regime. For the largest particles (OM:180-100) as well as for the smallest
382
(OM:32-18), the first group (78≤m/z ≤170) shows a higher intensity. However, for samplesOM:100-56
383
andOM:56-32, the second group (m/z≥178) exhibits a higher contribution. Within the first group, for all
384
the samples the base peak is located atm/z 128 and the group features fairly constant peak and intensity
385
Figure 6: L2MS spectra of four samples (particle size 180-18 nm) obtained in theOMengine conditions (optimal operation, medium load). The analyses were performed with an ionization wavelengthλi=266 nm to target aromatic compounds.
New figure
distributions. The second aromatic group shows a distinct behavior, where the base peak ism/z 252 for
386
samplesOM:180-100,OM:56-32, andOM:32-18, butm/z 228 for sampleOM:100-56. However, with
387
the exception of the peak atm/z 228, peak and intensity distributions change only slightly from sample to
388
sample. The change in the relative intensity of the m/z 228 peak is probably linked to a C18H+12 isomer
389
present on particles in the size range 100 – 32 nm. The fact that only species up to m/z 278 can be
390
seen on sampleOM:32-18 might be explained by an overall lower intensity, as this sample had the lowest
391
coverage of deposited particles. However, since the contribution from the highest masses (e.g. m/z≥275)
392
does not exceed a few percent of that given by the whole distribution of aromatic compounds making up
393
the second group (≥3-ring aromatic compounds), we can still derive the intensity ratio of species from the
394
second group to those belonging to the first one. A low ratio observed for this sample suggests that smaller
395
particles contain mostly light aromatic species.
396
The samples also contain a variety of aliphatic species, detected with a different ionization scheme
397
(λi=118nm)(Popovicheva et al., 2017), Fig. S3. Ion series with the formula CnH+2n−1 is typical of cy-
398
cloalkanes and alkenes while the CnH+2n+1 series (alkyl fragments) is typical of linear or branched alkane
399
compounds (McLafferty and Tureek, 1993). The series atm/z 67, 81, 95, and 109 (CnH2n−3 fragments of
400
bicycloalkanes) is also present, however with a very low intensity. The asymmetrical shape of the peaks is a
401
sign of fragmentation. The fragmentation pattern is characteristic of the presence of aliphatic hydrocarbons,
402
alkanes (CnH2n+2). As alkane cations are not stable, especially if the excess internal energy is high after
403
ionization, they can easily fragment. Since lubricating oils tend to be dominated by cycloalkanes due to
404
the deliberate removal ofn-alkanes during a de-waxing process (Tobias et al., 2001), an intense signal for
405
cycloalkanes is usually a sign of oil contribution. The main source of aliphatic compounds can be derived
406
from the ratio of n-alkanes to cycloalkanes. The signal corresponding to alkanes is superior to that of
407
neighboring cycloalkanes in the mass rangem/z = 67 – 71 and 81 – 85 when the fuel is the main source
408
of these compounds, and inferior for the lubricating oil (Dallmann et al., 2014; Sakurai et al., 2003; Tobias
409
et al., 2001). Ion signals at m/z 71 and 85 are lower than signals at m/z 69 and 83, suggesting that the
410
measured exhaust particles mainly consist of unburnt lubricating oil (in case of diesel particles this pattern
411
would be caused by at least 95% oil and 5% fuel, while in the case of gasoline, the ratio should be higher
412
(Dallmann et al., 2014; Sakurai et al., 2003). The one exception is observed for the sampleOM:100-56
413
where the signal atm/z 69 is lower than that of m/z 71 (r =S69/S71≈0.8), while the m/z 83 to 85 ratio
414
is close to unity. Since the source of organic species in the exhaust includes fuel, lubricating oil, and their
415
partially oxidized products, it is possible that particles containing the most unburnt fuel are concentrated
416
in the size bin associated with sampleOM:100-56(100 – 56 nm) and decreases towards both smaller and
417
larger particles. Peaks atm/z 50, 64, 66 and 78 can be attributed to fragments of aromatic species. Series
418
of highly unsaturated aliphatic compounds are present as well: C2nH+2 and C2nH+4. These series were at-
419
tributed to polyynic fragments, known to be present in rich premixed flames and play an important role in
420
the formation of combustion-generated particles (Hansen et al., 2008; Li et al., 2009).
421
To classify the samples and uncover differences and similarities between them, principal component
422
analysis (PCA) was used (Irimiea et al., 2018, 2019; Duca et al., 2019). This statistical method is able to
423
reduce the number of dimensions of complex mass-spectrometric data, thus increasing its readability, while
424
still preserving most of the original information. Before applying PCA, the data was first normalized and
425
baseline corrected (Irimiea et al., 2018, 2019; Duca et al., 2019) while all peaks associated with the substrate
426
were disregarded. Each principal component (PC) accounts for a defined percentage of the variance within
427
the dataset. For both ionization schemes, the first two components are responsible for most of the variation
428
between samples (containing particles of different sizes) (≈97 %), and therefore the number of dimensions
429
of the original data was reduced to only two. The contribution of mass peaks to each principal component
430
is represented by their loading (Fig. S4), which reflects the interpretation of individual components. To
431
extract the maximum amount of information, the data obtained with different ionization wavelengths was
432
treated separately. In case of the ionization at 266 nm, the first principal component (PC1) can be associated
433
with the total number of different aromatic species, since it receives a high negative contribution from both
434
alkylated and non-alkylated PAHs. The first group of aromatic compounds (78≤m/z≤170) shows a high
435
positive contribution to the second principal component (PC2) while the second group (m/z≥178) is related
436
to the negative PC2 value. For the data obtained with 118 nm ionization, PC1 is linked to all detected
437
aliphatic species and can be also seen as an indicator of the homogeneity of aliphatic signal throughout the
438
mass spectrum. PC2 has a strong contribution from highly unsaturated hydrocarbons (C2nH+2, C2nH+4,
439
positive PC2) and several aliphatic species (negative PC2), Fig. S4. Moreover, since contributions to
440
PC2 from C2nH+2, C2nH+4, and aliphatic species (CnH+2n−1, CnH+2n, CnH+2n+1) have the opposite sign, PC2
441
represents the ratio of highly unsaturated hydrocarbons to aliphatic species. In addition, the fact that peaks
442
atm/z 50, 52, 74 and 78 (fragments of aromatic species) have a high contribution to PC2 proves that these
443
species have the same origin, which is in agreement with soot formation models (Hansen et al., 2008; Li
444
et al., 2009).
445
Data in terms of PC1 and PC2 for both ionization wavelengths are presented in Fig. 7. There is a
446
notable separation between different samples, thus demonstrating the capability of the PCA method to
447
discriminate between samples that present, at first glance, similar mass spectra. Data points for each
448
sample are clustered together, proving that the sample surface is homogeneous and the reproducibility of
449
the method is high. SampleOM:180-100contains a large number of aromatic species (266 nm ionization),
450
with a higher contribution coming from the light-weight compounds (78≤m/z≤170). In contrast, sample
451
OM:100-56 exhibits a higher contribution from the second group of aromatic compounds (m/z≥178).
452
SampleOM:56-32seems to have fewer species, however heavier compounds (m/z≥178) still dominate. For
453
the smallest analyzed particles, i.e. sample OM:32-18, the contribution of lighter aromatic compounds
454
increases considerably, with only a few species detected from the second group. It appears that aromatic
455
species with a lower mass are present on all samples and are not bound to a specific particle size. In contrast,
456
there is a clear increase in the contribution from higher mass aromatic species forOM:100-56andOM:56-
457
32 samples, suggesting that these compounds preferably adsorb on particles in this size range (100 – 32
458
nm). SampleOM:100-56shows the most peaks coming from aliphatic species (118 nm ionization), including
459
peaks in the mass rangem/z 64 – 112 (which reflects as a negative PC1, the latter being representative of
460
all detected aliphatic species in Fig. 7). The smallest number of aliphatic compounds were detected on the
461
OM:56-32sample which also has a fairly inhomogeneous-looking spectrum (Fig. S3). The contribution of
462
highly unsaturated hydrocarbons (CnH+2 and CnH+4) decreases with the size of particles, being the highest
463